Hisashi Yashiro

The Non-hydrostatic Icosahedral Atmospheric Model: description and development

This article reviews the development of a global non-hydrostatic model, focusing on the pioneering research of the Non-hydrostatic Icosahedral Atmospheric Model (NICAM). Very high resolution global atmospheric circulation simulations with horizontal mesh spacing of approximately O (km) were conducted using recently developed supercomputers. These types of simulations were conducted with a specifically designed atmospheric global model based on a quasi-uniform grid mesh structure and a non-hydrostatic equation system. This review describes the development of each dynamical and physical component of NICAM, the assimilation strategy and its related models, and provides a scientific overview of NICAM studies conducted to date.

Madden–Julian Oscillation prediction skill of a new-generation global model demonstrated using a supercomputer

Global cloud/cloud system-resolving models are perceived to perform well in the prediction of the Madden–Julian Oscillation (MJO), a huge eastward -propagating atmospheric pulse that dominates intraseasonal variation of the tropics and affects the entire globe. However, owing to model complexity, detailed analysis is limited by computational power. Here we carry out a simulation series using a recently developed supercomputer, which enables the statistical evaluation of the MJO prediction skill of a costly new-generation model in a manner similar to operational forecast models. We estimate the current MJO predictability of the model as 27 days by conducting simulations including all winter MJO cases identified during 2003–2012. The simulated precipitation patterns associated with different MJO phases compare well with observations. An MJO case captured in a recent intensive observation is also well reproduced. Our results reveal that the global cloud-resolving approach is effective in understanding the MJO and in providing month-long tropical forecasts.

Impacts of cloud microphysics on trade wind cumulus: which cloud microphysics processes contribute to the diversity in a large eddy simulation?

This study investigated the impact of several cloud microphysical schemes on the trade wind cumulus in the large eddy simulation model. To highlight the differences due to the cloud microphysical component, we developed a fully compressible large eddy simulation model, which excluded the implicit scheme and approximations as much as possible. The three microphysical schemes, the one-moment bulk, two-moment bulk, and spectral bin schemes were used for sensitivity experiments in which the other components were fixed. Our new large eddy simulation model using a spectral bin scheme successfully reproduced trade wind cumuli, and reliable model performance was confirmed. Results of the sensitivity experiments indicated that precipitation simulated by the one-moment bulk scheme started earlier, and its total amount was larger than that of the other models. By contrast, precipitation simulated by the two-moment scheme started late, and its total amount was small. These results support those of a previous study. The analyses revealed that the expression of two processes, (1) the generation of cloud particles and (2) the conversion from small droplets to raindrops, were crucial to the results. The fast conversion from cloud to rain and the large amount of newly generated cloud particles at the cloud base led to evaporative cooling and subsequent stabilization in the sub-cloud layer. The latent heat released at higher layers by the condensation of cloud particles resulted in the development of the boundary layer top height.

Unrealistically pristine air in the Arctic produced by current global scale models

Black carbon aerosol (BCA) in the Arctic has profound impacts on the global climate system through radiation processes. Despite its enormous impacts, current global scale models, powerful tools for estimating overall impact, tend to underestimate the levels of BCA in the Arctic over several seasons. Using a global aerosol transport simulation with a horizontal grid resolution of 3.5 km, we determined that a higher resolution significantly reduced the underestimation of BCA levels in the Arctic, mainly due to an enhancement of the representation of low-pressure and frontal systems. The BCA mass loading in the Arctic simulated with 3.5-km grid resolution was 4.2-times larger than that simulated with coarse (56-km) grid resolution. Our results also indicated that grid convergence had not occurred on both the contrast between the cloud/cloud free areas and the poleward BCA mass flux, despite the use of the 3.5-km grid resolution. These results suggest that a global aerosol transport simulation using kilometre-order or finer grid resolution is required for more accurate estimation of the distribution of pollutants in the Arctic.

Performance Analysis and Optimization of Nonhydrostatic ICosahedral Atmospheric Model (NICAM) on the K Computer and TSUBAME2.5

We summarize the optimization and performance evaluation of the Nonhydrostatic ICosahedral Atmospheric Model (NICAM) on two different types of supercomputers: the K computer and TSUBAME2.5. First, we evaluated and improved several kernels extracted from the model code on the K computer. We did not significantly change the loop and data ordering for sufficient usage of the features of the K computer, such as the hardware-aided thread barrier mechanism and the relatively high bandwidth of the memory, i.e., a 0.5 Byte/FLOP ratio. Loop optimizations and code cleaning for a reduction in memory transfer contributed to a speed-up of the model execution time. The sustained performance ratio of the main loop of the NICAM reached 0.87 PFLOPS with 81,920 nodes on the K computer. For GPU-based calculations, we applied OpenACC to the dynamical core of NICAM. The performance and scalability were evaluated using the TSUBAME2.5 supercomputer. We achieved good performance results, which showed efficient use of the memory throughput performance of the GPU as well as good weak scalability. A dry dynamical core experiment was carried out using 2560 GPUs, which achieved 60 TFLOPS of sustained performance.

Scalable rank-mapping algorithm for an icosahedral grid system on the massive parallel computer with a 3-D torus network

Abstract In this paper, we develop a rank-mapping algorithm for an icosahedral grid system on a massive parallel computer with the 3-D torus network topology, specifically on the K computer. Our aim is to improve the weak scaling performance of the point-to-point communications for exchanging grid-point values between adjacent grid regions on a sphere. We formulate a new rank-mapping algorithm to reduce the maximum number of hops for the point-to-point communications. We evaluate both the new algorithm and the standard ones on the K computer, using the communication kernel of the Nonhydrostatic Icosahedral Atmospheric Model (NICAM), a global atmospheric model with an icosahedral grid system. We confirm that, unlike the standard algorithms, the new one achieves almost perfect performance in the weak scaling on the K computer, even for 10,240 nodes. Results of additional experiments imply that the high scalability of the new rank-mapping algorithm on the K computer is achieved by reducing network congestion in the links between adjacent nodes.

Automatic generation of efficient codes from mathematical descriptions of stencil computation

Programming in HPC is a tedious work. Therefore functional programming languages that generate HPC programs have been proposed. However, they are not widely used by application scientists, because of learning barrier, and lack of demonstrated application performance. We have designed Formura which adopts application-friendly features such as typed rational array indices. Formura users can describe mathematical concepts such as operation over derivative operators using functional programming. Formura allows intuitive expression over array elements while ensuring the program is a stencil computation, so that state-of-the-art stencil optimization techniques such as temporal blocking is always applied to Formura-generated program. We demonstrate the usefulness of Formura by implementing a preliminary below-ground biology simulation. Optimized C-code are generated from 672 bytes of Formura program. The simulation was executed on the full nodes of the K computer, with 1.184 Pflops, 11.62% floating-point-instruction efficiency, and 31.26% memory throughput efficiency.

Single Precision in the Dynamical Core of a Nonhydrostatic Global Atmospheric Model: Evaluation Using a Baroclinic Wave Test Case

AbstractReducing the computational cost of weather and climate simulations would lower electric energy consumption. From the standpoint of reducing costs, the use of reduced precision arithmetic has become an active area of research. Here the impact of using single precision arithmetic on simulation accuracy is examined by conducting Jablonowski and Williamson’s baroclinic wave tests using the dynamical core of a global fully compressible nonhydrostatic model. The model employs a finite volume method discretized on an icosahedral grid system and its mesh size is set to 220 km, 56 km, 14 km, and 3.5 km. When double precision arithmetic is fully replaced by single precision arithmetic, a spurious wavenumber-5 structure becomes dominant in both hemispheres, rather than the expected baroclinic wave growth only in the northern hemisphere. It was found that this spurious wave growth comes from errors in the calculation of grid cell geometrics. Therefore we performed an additional simulation using double precisi...

CONeP: A cost-effective online nesting procedure for regional atmospheric models

Abstract We propose a cost-effective online nesting procedure (CONeP) for regional atmospheric models to improve computational efficiency. The conventional procedure of online nesting is ineffective because computations are executed sequentially for each domain, and it does not enable users freely to determine the number of computational nodes. However, CONeP can completely avoid this limitation through three actions: 1) splitting the processes into multiple subgroups; 2) making each subgroup manage just one domain; and 3) executing the computations for each domain simultaneously. Since users can assign an optimal number of nodes to each domain, the model with CONeP is computationally efficient. We demonstrate the computational advantage of CONeP over the conventional procedure, comparing the elapsed times with both procedures on a supercomputer. The elapsed time with CONeP is markedly shorter than that observed with the conventional procedure using the same number of computational nodes. This advantage becomes more significant as the number of nesting domains increases.

Validation of high-resolution aerosol optical thickness simulated by a global non-hydrostatic model against remote sensing measurements

A high-performance computing resource allows us to conduct numerical simulations with a horizontal grid spacing that is sufficiently high to resolve cloud systems. The cutting-edge computational capability, which was provided by the K computer at RIKEN in Japan, enabled the authors to perform long-term, global simulations of air pollutions and clouds with unprecedentedly high horizontal resolutions. In this study, a next generation model capable of simulating global air pollutions with O(10 km) grid spacing by coupling an atmospheric chemistry model to the Non-hydrostatic Icosahedral Atmospheric Model (NICAM) was performed. Using the newly developed model, month-long simulations for July were conducted with 14 km grid spacing on the K computer. Regarding the global distributions of aerosol optical thickness (AOT), it was found that the correlation coefficient (CC) between the simulation and AERONET measurements was approximately 0.7, and the normalized mean bias was -10%. The simulated AOT was also compar...